|
|
|
|
|
|
|
generate structures which perform well (are fit) in the particular environment confronting it, and it must do this efficiently. Interest centers on robust adaptive plansplans which are efficient over the range of environments e they may encounter. Giving robustness precise definition and discovering something of the factors which make an adaptive plan robust is the formal distillation of questions about efficiency. Because efficiency is critical, the study of robustness has a central place in the formal development. |
|
|
|
|
|
|
|
|
The discussion of genetic systems emphasized two general requirements bearing directly on robustness: (1) The adaptive plan must retain advances already made, along with portions of the history of previous plan-environment interactions. (2) The plan must use the retained history to increase the proportion of fit structures generated as the overall history lengthens. The same discussion also indicated the potential of a particular class of adaptive plansthe reproductive plans. One of the first tasks, after setting out the formal framework, will be to provide a general definition of this class of plans. Lifting the reproductive plans from the specific genetic context makes them useful across the full spectrum of fields in which adaptation has a role. This widened role for reproductive plans can be looked upon as a first validation of the formalism. A much more substantial validation follows closely upon the definition, when the general robustness of reproductive plans is proved via the formalism. Later we will see how reproductive plans using generalized genetic operators retain and exploit their histories. Throughout the development, reproductive plans using genetic operators will serve to illuminate key features of adaptation and, in the process, we will learn more of the robustness, wide applicability, and general sophistication of such plans. |
|
|
|
|
|
|
|
|
Summarizing: This entire survey has been organized around the concept of an adaptive plan. The adaptive plan, progressively modifying structure by means of suitable operators, determines what structures are produced in response to the environment. The set of operators W and the domain of action of the adaptive plan  (i.e., the attainable structures) determine the plan's options; the plan's objective is to produce structures which perform well in the environment E confronting it. The plan's initial uncertainty about the environmentits room for improvementis reflected in the range of environments e in which it may have to act. The related performance measures µE,  , change from environment to environment since the same structure performs differently in different environments. These objects lie at the center of the formal framework set out in chapter 2. Chapter 3 provides illustrations of the framework as applied to genetics, economics, game-playing, searches, pattern recognition, statistical inference, control, function optimization, and the central nervous system. |
|
|
|
|
|